Blind Equalization Algorithms Based On Wavelet Embedded Neural Networks

The blind equalize technology does not need to transmit the training sequence to be possible to track the channel change, and enhances the data effectively the transmission speed and the reliability. However, the severe inter-symbol interference, caused by multi-path effects and channel distortion of sound propagation, degrades the reliability and decreases the rate of the propagation, is the chief obstacle of high speed underwater sound correspondence. Along with the development modernized from underwater sound correspondence to the high speed intelligence transmission, and it has the important meaning to analysis and research the theory and the algorithm of blind equalize in view of the underwater sound channel.Using wavelet transform and neural network as analyzing tool, the article has conducted the thorough research to the structure of neural network equalizer and the algorithm of blind equalization The main contribution of this dissertation are summarized as follows:1 Analyzing blind equalization algorithms based on Feed-forward Neural Network(1) The blind equalization Feed-forward Neural Network algorithm (FNN) based on momentum (MFNN) is proposed. Via analyzing the bases of the characteristic of underwater sound channel and the structure of neural network equalizer, induced the algorithm of momentum, the proposed algorithm can improve the character of the traditional FNN.(2) The Feed-forward Neural Network based on Super-Exponential Iterative (SEI-FNN) is proposed. After analyzing the algorithm of SEI, the proposed algorithm has ability to improve convergence rate and to reduce steady error via full using the whiten ability of SEI. The efficiency of the proposed algorithm is proved by computer simulation in underwater acoustic channels.2 Analyzing combined blind equalization feed-forward neural network algorithms based on orthogonal wavelet transform(1) The feed-forward neural network blind equalization algorithm based on orthogonal wavelet transform (OWT-FNN) is proposed. Induced the orthogonal wavelet transform into the neural network equalizer, the autocorrelation of the neural network input signal various components can be effectively reduced after carried on Orthogonal Wavelet Transform (WT) so as to improve the equalization effect. The result of underwater sound channel's computer simulation has proven the good performance of this algorithm.(2) A blind equalization algorithm of neural network based on orthognal wavelet transform fractionally spaced (T/4-FSE-WT-FNN) is proposed. after analyzing the basis of FSE and FNN, aiming at the lower convergence rate and the bigger steady error of the FNN, integrating the two algorithms and improving the character of the proposed algorithm. The result of underwater sound channel's computer simulation has proven the good performance of this algorithm.3 Analyzing blind equalization feed-forward neural network algorithm based on orthogonal wavelet packet transformThe feed-forward neural network blind equalization algorithm based orthogonal wavelet packet transform (OWPT-FNN) is proposed. The autocorrelation of input signal is further reduced via the signal scale space and wavelet space are decomposed respectively by orthogonal wavelet packet transform. Hence, we integrated WPT with blind equalization algorithm of neural network; OWPT-FNN has faster convergence rate and lower steady error. The result of underwater sound channel's computer simulation has proved the good performance of this algorithm.4 Analyzing the blind equalization based on wavelet embedded neural networkUsing wavelet function as the transfer function of neural network, the neural network weight vector of output and input layers can be adjusted by the iterative process of scale factor and displacement factor. And the character can be optimized.(1) The wavelet neural network equalization algorithm based on spatial diversity (SDE-WNN) is proposed. Introduced the spatial diversity into wavelet neural network (WNN), employed spatial diversity for reducing fading effect, it can speed up the convergence rate and reduced the steady error. The sparse underwater sound channel's computer simulation has proved the good performance of this algorithm.(2) AT/2 fractionally spaced based on wavelet neural network blind equalization algorithm (T/2-FSE-WNN) is proposed, Analyzed the bigger sampling rate of received signals of fractionally spaced equalizer (FSE) and approximation capability of wavelet neural network, two sub-channels are used in the wavelet neural network equalizer, the merits of FSE and WNN are integrated. Through simulation comparison to WNN and T/2-FSE-FNN, confirmed the propose algorithm to speed up the convergence rate and reduce the remainder steady error and root mean square error. Aiming at QAM signal, the proposed algorithm has the character of carrier recovery.